Short-Term Load Forecasting for Virtual Power Plants Using Time Series Analysis and Open Energy Data

Short-term load forecasting is helpful and useful in the management and operation of Virtual Power Plants, management of distributed renewable energy sources and stability of the power grid. This paper presents a dual-model time series forecasting approach where different models like ARIMA, Prophet, and LSTM have been introduced for forecasting the aggregated load of VPPs accurately. In the work, datasets of open energy consumption are used to train and test the proposed models; such datasets reflect various and often quite intricate patterns of consumption. The feature selection combines the merits of statistical analysis and machine learning providing near-optimal results for modelling of seasonality, trends and other non-linear relationships in load data. These real-life scenarios make an important aspect of the analysis in order to evaluate its effectiveness with varying penetration levels of renewable energy sources and consumer demand matrices. Such improvement of forecasting accuracy and its enhancement can be considered as valuable for VPP operators and stimulate further development of smart grid technologies. The results highlights the applicability of fusing output from various time series models and the use of open source energy data in improving short-term load forecasting of VPP thus resulting in a more robust energy system.